OpenSeeker-v2-30B-SFT is a 30B-scale search agent trained with supervised fine-tuning (SFT) on informative and high-difficulty search trajectories. It is released as part of OpenSeeker-v2: Pushing the Limits of Search Agents with Informative and High-Difficulty Trajectories.
Highlights
Deep search capabilities have become an indispensable competency for frontier Large Language Model (LLM) agents, yet their development remains dominated by industrial giants. The typical industry recipe involves a highly resource-intensive pipeline spanning pre-training, continual pre-training (CPT), supervised fine-tuning (SFT), and reinforcement learning (RL).
In this report, we show that when fueled with informative and high-difficulty trajectories, a simple SFT approach can be surprisingly powerful for training frontier search agents. By introducing three simple data synthesis modifications, we establish a stronger baseline:
- Scaling knowledge graph size for richer exploration.
- Expanding the tool set size for broader functionality.
- Applying strict low-step filtering to keep high-difficulty trajectories.
Trained on merely 10.6k data points, OpenSeeker-v2 achieves state-of-the-art performance across four benchmarks among 30B-sized agents with the ReAct paradigm:
- 46.0% on BrowseComp.
- 58.1% on BrowseComp-ZH.
- 34.6% on Humanity's Last Exam.
- 78.0% on xbench.
OpenSeeker-v2 surpasses Tongyi DeepResearch, which is trained with a heavier CPT+SFT+RL pipeline and achieves 43.4%, 46.7%, 32.9%, and 75.0% on the same benchmarks, respectively.
Notably, OpenSeeker-v2 represents the first state-of-the-art search agent within its model scale and paradigm to be developed by a purely academic team using only SFT. We are excited to open-source the OpenSeeker-v2 model weights and share our simple yet effective findings to make frontier search agent research more accessible to the community.